A method and system for controlling feed fermentation parameters

Abstract

This invention relates to the field of enzyme catalysis technology and discloses a method and system for controlling feed fermentation parameters. One method includes: calculating a metabolome similarity index based on metabolite distribution; constructing an inter-enzyme interaction model based on an enzyme activity correlation matrix and enzyme activity; defining a fermentation efficiency function and solving the function to obtain the optimal parameter combination; calculating the parameter deviation between the current actual parameter values ​​and the optimal fermentation parameter combination; and calculating the amount of fermentation parameters that need to be adjusted based on the parameter deviation. This invention improves the overall efficiency of the fermentation process, shortens the fermentation cycle, and increases the yield through the construction of an inter-enzyme interaction network and comprehensive parameter control.

Description

Technical Field

[0001] This invention relates to the field of enzyme catalysis technology, and more specifically, to a method and system for controlling feed fermentation parameters. Background Technology

[0002] Feed fermentation technology is a crucial supporting technology for modern animal husbandry. Through microbial fermentation, it degrades anti-nutritional factors in raw materials, improving feed digestibility and nutritional value. Traditional feed fermentation processes rely heavily on manual experience for parameter control, adjusting fermentation conditions through periodic sampling and testing. This approach suffers from problems such as response lag and insufficient precision.

[0003] With the development of fermentation technology, modern feed fermentation widely adopts multi-enzyme complex fermentation systems. By adding various enzyme preparations such as cellulase, protease, amylase, and phytase, comprehensive modification of feed raw materials can be achieved. Compared with single-enzyme systems, this multi-enzyme complex system has higher conversion efficiency and better product quality, but it also brings a significant increase in control complexity.

[0004] Chinese Patent CN117229905A discloses a method and system for controlling fermentation of biological feed. Based on sensor technology and employing real-time data analysis algorithms, it detects enzyme activity in feed in real time, analyzes its trends, and generates a real-time enzyme activity data report. This invention accurately monitors and predicts enzyme activity in feed through real-time data analysis and deep learning prediction algorithms. An adaptive control algorithm automatically adds enzyme preparations, improving production efficiency and stability. Based on component analysis and machine learning correlation algorithms, it generates targeted enzyme preparation addition schemes, improving the effectiveness and specificity of enzyme preparations. Through an enzyme activity-environment response model and environmental parameter adjustment algorithms, it adjusts the microenvironment in real time to ensure the enzyme is in an optimal activity state. Bioinformatics tools and protein engineering techniques are used to fine-tune the enzyme structure and verify its stability, generating enzyme preparations with enhanced stability.

[0005] Existing fermentation parameter control technologies mainly employ independent parameter monitoring, focusing only on changes in the activity of a specific enzyme or the concentration of a certain type of metabolite. This method fails to reflect the impact of multi-enzyme interactions on the overall fermentation effect and ignores the correlation between metabolomic changes and enzyme activity changes; resulting in insufficient precision in fermentation process control, poor batch-to-batch product quality consistency, and low fermentation efficiency. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for controlling feed fermentation parameters in order to solve the above-mentioned problems.

[0007] This invention provides a method for controlling feed fermentation parameters, comprising the following steps:

[0008] Obtain the distribution of metabolites during feed fermentation and calculate the metabolome similarity index based on the distribution of metabolites;

[0009] Establish an enzyme activity correlation matrix, determine enzyme activity based on metabolite distribution, construct an enzyme interaction model based on the enzyme activity correlation matrix and enzyme activity, and output the enzyme activity of each enzyme in the future time period from the enzyme interaction model.

[0010] By combining the metabolome similarity index and the activity of each enzyme over a future period, a fermentation efficiency function is defined, and the optimal combination of fermentation parameters is obtained by solving the fermentation efficiency function.

[0011] Calculate the parameter deviation between the current actual parameter values ​​and the optimal combination of fermentation parameters, and calculate the amount of fermentation parameters that need to be adjusted based on the parameter deviation.

[0012] Furthermore, the enzyme activity correlation matrix represents a two-dimensional table of interactions between all enzymes, where each element represents the degree of influence of one enzyme on another, and the degree of influence is represented by an influence coefficient. The number of rows and columns in the enzyme activity correlation matrix is ​​equal to the total number of enzymes. Each element in the enzyme activity correlation matrix represents the influence coefficient of the enzyme in the row on the enzyme in the column. Positive values ​​indicate a promoting effect, negative values ​​indicate an inhibiting effect, and zero values ​​indicate no interaction.

[0013] Furthermore, the enzyme interaction model is a differential equation that describes the change of enzyme activity over time through mathematical expressions. The differential equation includes the growth of enzyme activity and the degradation of enzyme activity. The growth of enzyme activity is the product of the enzyme's intrinsic activity coefficient and the current enzyme activity, plus the sum of the products of the preset influence coefficient and the activities of each enzyme. The degradation of enzyme activity is represented by the product of the degradation rate and the current enzyme activity.

[0014] Furthermore, the fermentation efficiency function includes a weighted total enzyme activity term and a metabolomics similarity contribution term; the total enzyme activity term is the sum of the activities of each enzyme multiplied by the relative importance coefficient of each enzyme, and then multiplied by a preset total enzyme activity weighting coefficient; the metabolomics similarity contribution term is the negative value of the metabolomics similarity index multiplied by the preset metabolomics similarity weighting coefficient. Each enzyme activity is input into the fermentation efficiency function, and the fermentation efficiency function outputs the fermentation efficiency; where:

[0015] The relative importance coefficients of various enzymes are determined based on the effects of each enzyme on metabolites. The effect is the average percentage change in the distribution of the corresponding metabolite caused by increasing the activity of a certain enzyme within a preset range.

[0016] Furthermore, the fermentation efficiency function is solved using a particle swarm optimization (PSO) simulated annealing mixing algorithm, the steps of which include:

[0017] Multiple particles are randomly generated within the limits of fermentation parameters. Each particle contains a fermentation parameter, which includes temperature, pH value and ionic strength. The limits of fermentation parameters are the range between the maximum and minimum values ​​of the fermentation parameters.

[0018] Calculate the fermentation efficiency value for each particle;

[0019] The particle updates its velocity and position based on its own optimal position and the global optimal position;

[0020] The acceptance probability is set, and the candidate solution is found by local search of the particle swarm using the simulated annealing algorithm. The acceptance probability is adjusted according to the annealing temperature and the efficiency difference, which is the difference between the target efficiency and the actual efficiency. The candidate solution is the combination of fermentation parameters that maximizes the overall fermentation efficiency. The adjustment rule for the acceptance probability is that when the efficiency difference is negative, the acceptance probability is 1, and when the efficiency difference is positive, the acceptance probability is the negative efficiency difference of the natural constant divided by the power of the current annealing temperature.

[0021] Based on the current solution, Gaussian noise with a mean of 0 and a standard deviation that is a preset multiple of the parameter range is added to each fermentation parameter to generate new fermentation parameters; where the current solution is the fermentation parameter in the particle currently being evaluated, and the parameter range is the difference between the maximum and minimum values ​​of the fermentation parameter;

[0022] The process stops when the maximum number of iterations is reached, or when the efficiency difference falls below the set iteration threshold after a preset number of iterations. The optimal solution is then output, which is the combination of fermentation parameters that maximizes the overall fermentation efficiency.

[0023] Furthermore, the method for calculating the fermentation efficiency value corresponding to each particle includes:

[0024] The environmental parameters are obtained by combining the temperature, pH value, and ionic strength of each particle. The environmental parameters, the current actual enzyme activity value, and the substrate concentration are input into the trained environmental impact model. The environmental impact model outputs the enzyme activity values. The enzyme activity values ​​are input into the enzyme interaction model to obtain the enzyme activity over a future period. The enzyme activity is input into the fermentation efficiency function, and the fermentation efficiency function outputs the fermentation efficiency value.

[0025] Furthermore, the current actual parameter values ​​are the actual fermentation parameter values ​​for the current fermentation process, including the actual temperature, actual pH value, and actual ionic strength;

[0026] The parameter deviation is the difference between the optimal combination of fermentation parameters and the actual fermentation parameter values, including temperature deviation, pH deviation, and ionic strength deviation. The temperature deviation is equal to the optimal temperature minus the actual temperature; the pH deviation is equal to the optimal pH value minus the actual pH value; and the ionic strength deviation is equal to the optimal ionic strength minus the actual ionic strength.

[0027] Furthermore, the fermentation parameters that need to be adjusted include temperature adjustment, pH adjustment, and ionic strength adjustment, among which:

[0028] The temperature adjustment amount equals the preset temperature proportional coefficient multiplied by the temperature deviation amount, plus the preset temperature integral coefficient multiplied by the time integral of the temperature deviation amount, plus the preset temperature differential coefficient multiplied by the time differential of the temperature deviation amount.

[0029] The pH adjustment amount equals the preset pH proportional coefficient multiplied by the pH deviation amount, plus the preset pH integral coefficient multiplied by the pH deviation amount over time, plus the preset pH derivative coefficient multiplied by the pH deviation amount over time.

[0030] The ion intensity adjustment is equal to the preset ion intensity proportional coefficient multiplied by the ion intensity deviation, plus the preset ion intensity integral coefficient multiplied by the time integral of the ion intensity deviation, plus the preset ion intensity differential coefficient multiplied by the time differential of the ion intensity deviation.

[0031] Furthermore, the method for calculating the metabolomics similarity index includes: obtaining the concentration of each metabolite in the distribution of metabolites; extracting the reference concentration and weight coefficient of each metabolite from the database; calculating the absolute difference between the concentration of each metabolite and the reference concentration, and dividing the absolute difference by the reference concentration to obtain the relative deviation; multiplying the relative deviation by the weight coefficient of the corresponding metabolite to obtain the weighted deviation; and summing the weighted deviations of multiple metabolites to obtain the metabolomics similarity index.

[0032] This invention provides a feed fermentation parameter control system for executing a feed fermentation parameter control method described above, the system comprising:

[0033] The identification module obtains the distribution of metabolites during the feed fermentation process and calculates the metabolome similarity index based on the distribution of metabolites.

[0034] The analysis module establishes an enzyme activity correlation matrix, determines enzyme activity based on the distribution of metabolites, constructs an enzyme interaction model based on the enzyme activity correlation matrix and enzyme activity, and outputs the activity of each enzyme in the future period of time from the enzyme interaction model.

[0035] The optimization module combines the metabolome similarity index and the activity of each enzyme over a future period to define a fermentation efficiency function, and solves the fermentation efficiency function to obtain the optimal combination of fermentation parameters.

[0036] The control module calculates the parameter deviation between the current actual parameter values ​​and the optimal fermentation parameter combination, and calculates the amount of fermentation parameters that need to be adjusted based on the parameter deviation.

[0037] The beneficial effects of this invention are as follows: This invention achieves precise control of the fermentation process through the synergistic analysis of metabolomics similarity index and enzyme interaction model, significantly improving product quality consistency compared to traditional single-parameter control methods; it quantifies the mutual influence between different enzymes through the enzyme interaction model, overcoming the limitation of traditional methods where single enzyme activity monitoring cannot reflect the overall fermentation effect; it uses a particle swarm optimization-simulated annealing hybrid algorithm to solve for the optimal fermentation parameter combination, avoiding the blindness and inefficiency of traditional trial-and-error methods; it establishes an environmental influence model, enabling real-time adjustment of the control strategy based on changes in environmental parameters; and it employs a PID control algorithm combined with a first-order filter, resulting in fast response and strong stability. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating a method for controlling feed fermentation parameters according to the present invention.

[0039] Figure 2 This is an example diagram of the fermentation efficiency function of a feed fermentation parameter control method according to the present invention;

[0040] Figure 3 This is a flowchart illustrating a multi-center adaptive control method according to the present invention;

[0041] Figure 4 This is a module example diagram of a feed fermentation parameter control system according to the present invention. Detailed Implementation

[0042] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0043] A method and system for controlling feed fermentation parameters includes the following embodiments:

[0044] Example 1

[0045] This embodiment provides a method for controlling feed fermentation parameters to address the problem that single enzyme activity monitoring in multi-enzyme complex fermentation systems cannot reflect the overall fermentation effect. Figure 1 As shown.

[0046] In this embodiment, the method is applied to an industrial feed fermentation production line equipped with multiple sensor arrays, metabolomics analysis equipment, a multi-channel fermentation parameter control system, and edge computing devices. The information acquired by this method includes real-time monitoring data of multi-enzyme activity, metabolite spectra during fermentation, spectral analysis results of feed ingredient components, real-time data of environmental parameters (temperature, pH, oxygen concentration, ionic strength), and a database of inter-enzyme interaction relationships.

[0047] The specific steps include:

[0048] Step 101: Online near-infrared spectroscopy analysis combined with a multi-parameter sensor array is used to obtain the distribution of metabolites during the fermentation process. The metabolic state of the fermentation process is evaluated by calculating the metabolome similarity index through the distribution of metabolites, providing data support for subsequent enzyme interaction analysis and fermentation parameter optimization.

[0049] Metabolite distribution refers to the concentration distribution of various metabolites at different spatial locations and time points within a fermentation vessel, including spatial distribution characteristics and temporal evolution characteristics. Spatial distribution characteristics are obtained using near-infrared spectroscopy probes installed at multiple points, reflecting the uniformity of the fermentation process and the mass transfer effect. Temporal evolution characteristics record the trajectory of the concentration changes of each metabolite over fermentation time, reflecting the kinetics of the fermentation reaction.

[0050] The specific steps for calculating the metabolomics similarity index include: obtaining the concentration of each metabolite in the metabolite distribution; extracting the reference concentration and weighting coefficient of each metabolite from the database; calculating the absolute difference between the concentration of each metabolite and the reference concentration, and then dividing by the reference concentration to obtain the relative deviation; multiplying the relative deviation by the weighting coefficient of the corresponding metabolite to obtain the weighted deviation; summing the weighted deviations of multiple metabolites to obtain the final metabolomics similarity index; the relative deviation can eliminate the influence of differences in the concentration magnitude of different metabolites; the weighting coefficient can reflect the differences in the importance of different metabolites to fermentation quality; and the use of the absolute difference avoids the problem of positive and negative deviations canceling each other out.

[0051] The types and quantities of metabolites were determined through prior laboratory analysis. A typical feed fermentation process monitors 20-30 key metabolites, such as lactic acid, propionic acid, and amino acids. Weighting coefficients were determined from historical production data using principal component analysis; for example, the weighting coefficient for lactic acid is 0.15-0.20, indicating its significant impact on fermentation quality. Reference concentrations were obtained by analyzing the concentrations of metabolites from historical high-quality batches; for example, the ideal concentration for lactic acid is 15-20 g / L.

[0052] The specific steps for determining weight coefficients using principal component analysis include: collecting metabolite concentrations and corresponding product quality scores from 100-200 fermentation batches in a historical database; standardizing the data to eliminate the influence of dimensions; calculating the covariance matrix and solving for eigenvalues ​​and eigenvectors; selecting principal components with a cumulative contribution rate of 85% or higher; and determining the weight of each metabolite based on its loading coefficient in the principal components. For example, lactic acid, which has a high loading coefficient in the first principal component, receives a larger weight of 0.18, while some amino acids with lower loading coefficients have weights of only 0.03-0.05.

[0053] The definition criteria for historically high-quality batches include: product protein digestibility ≥ 85% and harmful bacteria count < 10. 3 CFU / g and nutrient loss rate <5%. Historical batches meeting these criteria were screened, and the average concentration of each metabolite was calculated as a reference concentration. For example, in high-quality batches, lactic acid concentrations ranged from 15-20 g / L, with an average of approximately 17.5 g / L; therefore, 17.5 g / L was set as the reference concentration for lactic acid.

[0054] The weighting coefficients were determined from historical production data using principal component analysis. In this embodiment, the specific implementation steps of principal component analysis include: collecting metabolite concentration data from 500 production batches over the past 12 months; constructing a metabolite concentration matrix, where rows represent different batches, columns represent different metabolites, and each value in the matrix represents the concentration of a different metabolite from a different batch; standardizing the constructed metabolite concentration matrix to eliminate the influence of dimensions; calculating the covariance matrix of the constructed metabolite concentration matrix and solving for eigenvalues ​​and eigenvectors; selecting the top principal components with a cumulative contribution rate of 85% or higher. The cumulative contribution rate is calculated as follows: first, all eigenvalues ​​are sorted from largest to smallest; then, the proportion of each eigenvalue to the total eigenvalues ​​is calculated, and this proportion represents the contribution rate of the principal component; finally, the contribution rates of each principal component are summed sequentially. Selecting the top principal components with a cumulative contribution rate of 85% or higher is based on a balance between information retention and computational efficiency. The 85% threshold indicates that the selected principal components can explain 85% of the variation information in the original data, preserving the main data features while achieving effective dimensionality reduction. By analyzing the loading coefficients of each metabolite in the selected principal components, their contribution to fermentation quality is determined and normalized into weighting coefficients. The loading coefficient represents the correlation between the original metabolite variables and the principal components. The larger the absolute value of the loading coefficient, the greater the contribution of the metabolite to the principal components, and thus the more important its impact on fermentation quality.

[0055] The reference concentration was obtained by analyzing the metabolite concentrations of historically high-quality batches. Historically high-quality batches are defined as production batches that simultaneously meet the following criteria: crude protein content of ≥18%, crude fiber content of ≤5%, digestibility of ≥85%, absence of mold and off-odors, and compliance with national microbiological standards. From the past 24 months of production records, 150 high-quality batches meeting these criteria were selected. Metabolite concentration data for these batches during the fermentation stabilization period (60%-80% of the total fermentation time) were extracted. The mean and standard deviation of each metabolite concentration were calculated, and the range of mean ± 0.5 times the standard deviation was used as the reference concentration range for that metabolite. This range was set based on statistical principles, covering approximately 38% of the data distribution, ensuring the representativeness of the reference concentration while avoiding an overly broad range that could lead to decreased control precision.

[0056] Online near-infrared spectroscopy analysis technology refers to the use of a near-infrared spectrometer to detect the characteristic absorption peaks of organic molecules in fermentation broth in real time, and to infer the concentration changes of major metabolites through spectral data analysis. This technology uses a fiber optic probe with a diameter of 6-10 mm, which has good corrosion resistance and anti-fouling capabilities, and can operate stably for a long time in the fermentation environment. The detection wavelength range is 800-2500 nm, covering the characteristic absorption region of most organic compounds.

[0057] The multi-parameter sensor array includes devices such as an online pH meter, dissolved oxygen electrode, conductivity sensor, and temperature sensor, which are used to supplement ionic metabolites and environmental parameters that cannot be detected by near-infrared spectroscopy.

[0058] The distribution of metabolites was acquired through a combination of online near-infrared spectroscopy and a multi-parameter sensor array. Five to eight sampling points were set at different heights and radial positions within the fermentation vessel. Each sampling point was equipped with a near-infrared spectroscopy probe and corresponding auxiliary sensors, enabling simultaneous detection of concentration changes in 15 to 20 key metabolites. The sampling frequency was set to once every 30 minutes to ensure the capture of dynamic changes in metabolite concentrations during fermentation.

[0059] The selection of key metabolites is based on the biochemical reaction pathway analysis of feed fermentation, mainly including: organic acids (lactic acid, propionic acid, butyric acid, acetic acid), soluble sugars (glucose, fructose, sucrose), soluble proteins, ammonia nitrogen, volatile fatty acids, and other important metabolic indicators. The selection criteria for these metabolites include: significant concentration changes during fermentation, important impact on product quality, ease of detection by near-infrared spectroscopy, and distinct spectral characteristics.

[0060] Near-infrared spectroscopy is based on the absorption of overtones and combination frequencies of molecular vibrations. Organic acids exhibit characteristic absorptions in the 1400-1500 nm and 1900-2000 nm regions for their CH and OH bonds; carbohydrates show significant absorption in the 1200-1300 nm region for their CH bonds; and proteins have characteristic peaks in the 1500-1600 nm and 2100-2200 nm regions for their NH and CH bonds.

[0061] The 30-minute sampling frequency was determined based on fermentation kinetics analysis and equipment response time. Analysis of the time constants of typical feed fermentation processes revealed that the significant concentration changes of most metabolites occur within a 1-2 hour period; therefore, the 30-minute sampling interval effectively captures important concentration trends. Furthermore, the near-infrared spectrometer's single detection time is 2-3 minutes, providing ample time for data processing and quality control.

[0062] Step 102: After obtaining metabolomics features, it is necessary to understand the interrelationships between various enzymes in the fermentation system. Therefore, an enzyme activity correlation matrix is ​​established, which represents the mutual influence between different enzymes. An enzyme interaction model is constructed based on the enzyme activity correlation matrix, and the enzyme interaction model outputs the activity of each enzyme in the future.

[0063] An enzyme activity correlation matrix is ​​a two-dimensional table representing the interactions between all enzymes, where each element represents the degree of influence of one enzyme on another. The number of rows and columns in the matrix equals the total number of enzymes. Each element in the enzyme activity correlation matrix represents the influence coefficient of the enzyme in the corresponding row on the enzyme in the corresponding column; positive values ​​indicate a promoting effect, negative values ​​indicate an inhibiting effect, and zero values ​​indicate no interaction.

[0064] Enzyme activity is measured using a specific substrate method, which involves designing a specific substrate for each enzyme and indirectly measuring enzyme activity by monitoring the substrate consumption rate or product formation rate through metabolite distribution. For example, cellulase activity is measured by monitoring the degradation rate of sodium carboxymethyl cellulose (CMC-Na), with units of U / mL (the amount of enzyme that catalyzes the production of 1 micromole of reducing sugar per minute); protease activity is measured by monitoring the hydrolysis rate of casein; and amylase activity is measured by monitoring the hydrolysis rate of soluble starch.

[0065] The total number of enzymes is determined through fermentation formulation and raw material analysis. A typical industrial feed fermentation process involves 5-8 key enzymes. Taking corn-soybean meal feed as an example, the main components include starch (40-50%), protein (18-22%), cellulose (3-5%), and fat (3-4%), thus requiring corresponding amylase, protease, cellulase, and lipase. In addition, phytase is needed to decompose phytic acid and improve phosphorus utilization; xylanase is needed to decompose hemicellulose and improve feed digestibility.

[0066] The influence coefficients were preset values ​​obtained through in vitro enzymological experiments and data mining methods. The in vitro enzymological experiments were designed as follows: single enzyme A and a combination of enzymes (including both enzyme A and enzyme B) were added to standard buffer. The enzymes were incubated at the same temperature and pH (default temperature: 37℃, default pH: 6.5). Enzyme activities were measured periodically, for example, after one hour. The rate of change in enzyme A activity was compared between the single-enzyme system and the two-enzyme system; this rate of change was the influence coefficient of enzyme B on enzyme A.

[0067] Based on the enzyme activity correlation matrix, an enzyme interaction model is constructed, which describes the change in the activity of each enzyme over time. The core idea of ​​the model is that the rate of change of the activity of each enzyme depends on its own intrinsic activity, the influence of other enzymes on it, and environmental degradation.

[0068] The mathematical expression of the enzyme interaction model describes the differential equation of enzyme activity changing over time. Specifically, the derivative of enzyme activity with respect to time equals the sum of two factors: one is the increase in enzyme activity, influenced by its inherent properties and the activities of other enzymes; the other is enzyme degradation. In the increase of enzyme activity, the product of the enzyme's inherent activity coefficient and its current activity represents the base growth rate, while the influence of other enzymes is represented by the sum of the products of their influence coefficients and the activities of each enzyme. Enzyme degradation is simply represented by the product of the degradation rate and the current enzyme activity, indicating natural enzyme degradation. This model provides a theoretical basis for optimizing fermentation processes and can be used to predict the dynamic changes in enzyme activity.

[0069] Intrinsic activity coefficient and degradation rate are two key parameters describing the changes in the activity of a single enzyme. The intrinsic activity coefficient represents the rate of increase in enzyme activity under ideal conditions, while the degradation rate represents the rate of enzyme inactivation in the environment. These two parameters are determined through stability experiments of a single enzyme. The specific methods include: measuring the initial activity of the enzyme under controlled conditions; sampling and measuring enzyme activity at different time points; and obtaining the intrinsic activity coefficient and degradation rate of the enzyme through exponential fitting.

[0070] The standard conditions for single enzyme stability experiments include a temperature of 37±0.5℃, pH of 6.5±0.1, ionic strength of 0.1 mol / L, and no interference from other enzymes. Ionic strength refers to half the sum of the products of the concentrations of all ions in the solution and their squares, used to characterize the electrolyte concentration of the solution, which affects the conformational stability of the enzyme and the charge distribution of the active site. The experimental period is 24 hours, with enzyme activity measured every 2 hours. The mathematical model for activity growth and decay is an exponential function; the change in enzyme activity over time is equal to the initial activity multiplied by the exponent of the natural constant, where the exponent is the intrinsic activity coefficient minus the degradation rate multiplied by time. The values ​​of the intrinsic activity coefficient and degradation rate are determined through nonlinear fitting. For example, the intrinsic activity coefficient of cellulase is 0.05–0.08 per hour, and the degradation rate is 0.01–0.02 per hour, reflecting its activity change characteristics under fermentation conditions.

[0071] The numerical solution of the differential equations was performed using the fourth-order Runge-Kutta method, which offers high accuracy and stability. A computational step size of 0.1 hours was set to ensure both accuracy and control of computational load. The initial conditions were the activity values ​​of each enzyme at the start of fermentation.

[0072] The steps for implementing an enzyme interaction model include: collecting current enzyme activity data as the initial state of the model; solving differential equations using numerical integration methods (such as the Runge-Kutta method) to obtain the changing trends of enzyme activities over a future period; assessing the stability of the current fermentation state based on the prediction results; and providing a basis for decision-making in the next step of fermentation parameter optimization.

[0073] The prediction window for the future period is set to 4-6 hours. This time length is determined based on the characteristic time of the fermentation process. A prediction window that is too short cannot reflect the trend of enzyme activity changes, while a prediction window that is too long will reduce the prediction accuracy due to the accumulation of model errors. The trend of enzyme activity changes is the enzyme activity at each time interval within the future period, and the default value of the time interval is 30 minutes.

[0074] Model validation was conducted by comparing predicted and measured values, using the mean absolute percentage error (MAE) as the evaluation metric. Predicted values ​​refer to the theoretical values ​​of each enzyme activity for the next 4-6 hours, calculated using the enzyme interaction model, specifically obtained by numerically solving a system of differential equations using the fourth-order Runge-Kutta method. Measured values ​​refer to the actual enzyme activity data measured at the corresponding time points using the specific substrate method, including the activity values ​​of key enzymes such as cellulase, protease, and amylase, with measurements taken every 2 hours.

[0075] For each enzyme, the absolute value of the difference between the predicted value and the measured value is divided by the measured value, and then multiplied by 100% to obtain the absolute percentage error of the individual enzyme; then the average absolute percentage error of all enzymes is calculated to obtain the average absolute percentage error of the overall model.

[0076] The specific steps for model validation include: predicting enzyme activity changes over the next 4-6 hours at key time points such as the 12th, 24th, and 36th hours of fermentation using prior data; collecting actual samples and measuring the activity of each enzyme after the prediction time window ends; calculating the deviation between the predicted and measured values, and statistically analyzing the mean absolute percentage error. When the mean absolute percentage error is less than 15%, the model prediction accuracy is considered acceptable. If the error exceeds 15%, the cause needs to be analyzed, which may include factors such as model parameter drift, changes in raw material composition, and fluctuations in environmental conditions.

[0077] In this way, the model can predict the dynamic changes in the activity of each enzyme under different combinations of enzymes, quantify the impact of enzyme interactions on the overall fermentation process, provide a theoretical basis for the optimization of fermentation parameters, and complement the metabolomics analysis results in step 101 to jointly constitute a comprehensive evaluation system for the fermentation process.

[0078] Step 103: Combining the output of the metabolomics similarity index and the enzyme interaction model, a fermentation efficiency function is defined, and the optimal combination of fermentation parameters is obtained by solving the fermentation efficiency function using a particle swarm optimization-simulated annealing hybrid algorithm. The analysis results of the first two steps are transformed into specific control strategies, realizing the key transformation from data analysis to fermentation parameter control.

[0079] Specific methods are as follows Figure 2 As shown, it includes:

[0080] The fermentation efficiency function is an evaluation index that comprehensively considers enzyme activity and metabolomics similarity index. This function consists of two parts: a weighted total enzyme activity term and a metabolomics similarity contribution term. The total enzyme activity term is the sum of the activities of each enzyme multiplied by its relative importance coefficient, and then multiplied by the total enzyme activity weight coefficient. The metabolomics similarity contribution term is the negative value of the metabolomics similarity index multiplied by the metabolomics similarity weight coefficient, because a smaller metabolomics similarity index indicates a more ideal state. The fermentation efficiency function outputs the fermentation efficiency by inputting the enzyme activities into it.

[0081] The fermentation efficiency function is designed based on multi-objective optimization theory, merging multiple evaluation indicators into a single objective function through weighted summation. The weight coefficients are determined using the analytic hierarchy process (AHP), with expert scoring and pairwise comparisons used to determine the relative importance of each indicator.

[0082] In practical applications, the weighting coefficient for total enzyme activity was set to 0.6-0.7, and the weighting coefficient for metabolome similarity was set to 0.3-0.4, reflecting that in the current fermentation system, enzyme activity contributes slightly more to the overall efficiency than metabolome status.

[0083] The relative importance coefficients of various enzymes are determined based on their effects on metabolites. To prevent nonlinear effects, the effect is the average percentage change in metabolite distribution caused by a fixed enzyme activity increasing within a preset range. The default value for the preset range is 1% to 30%. A lower limit of 1% ensures measurement accuracy, as enzyme activity changes below 1% are close to the detection error range, making it difficult to accurately quantify their impact on metabolites. An upper limit of 30% avoids entering the nonlinear region; when enzyme activity changes exceed 30%, nonlinear effects such as substrate saturation and product inhibition may occur in the enzymatic reaction, leading to distorted effect calculations. This range covers the typical fluctuation range of enzyme activity during fermentation, ensuring statistical significance while maintaining the validity of the linear assumption, providing a reliable basis for accurate calculation of the importance coefficients. The method for determining the effect is as follows:

[0084] Under standard fermentation conditions, keeping the activities of other enzymes constant, the activity of the target enzyme was increased by 1%, 5%, 10%, 15%, 20%, 25%, and 30%, respectively. The distribution of metabolites at each activity level was measured, and the percentage change in metabolite distribution was calculated. The arithmetic mean of these percentage changes was taken as the effect of the enzyme. The method for calculating the percentage change in metabolite distribution was to subtract the relative content of a certain metabolite in the control group from the relative content in the treatment group, divide by the relative content in the control group, and finally multiply by 100%.

[0085] Immobilized enzyme activity refers to the activity of all enzymes except the target enzyme remaining unchanged at a preset standard level during the effect determination process. The standard level is defined as the steady-state activity value of each enzyme under optimal fermentation conditions.

[0086] Fermentation parameters include temperature, pH, and ionic strength. Optimization of these parameters requires consideration of physical limitations, primarily the temperature range, pH range, and ionic strength range. These limitations are determined based on the activity characteristics of the enzyme preparation used and the stability of the feedstock, with typical values ​​of 25-55℃ temperature, 4.5-7.0 pH, and 0.05-0.2 mol / L ionic strength. These parameter limits are determined through enzymatic characterization studies and preliminary pilot-scale experiments to ensure that the fermentation parameters are within practically feasible ranges.

[0087] The determination of fermentation parameter limits is based on enzyme activity-temperature curve determination, enzyme activity-pH curve determination, determination of the effect of ionic strength on enzyme activity, feedstock stability testing, and equipment operating limitations. Among these, enzyme activity-temperature curve determination determines the optimal temperature range and inactivation temperature for each enzyme; enzyme activity-pH curve determination determines the optimal pH range and stable pH range for each enzyme; determination of the effect of ionic strength on enzyme activity determines the suitable range of ionic strength; feedstock stability testing determines the retention of nutrients under various fermentation parameters; and equipment operating limitations consider the operating range of heating, cooling, and pH adjustment equipment.

[0088] For example, cellulase has an optimal temperature of 45-50℃, begins to inactivate above 55℃, and its activity decreases significantly below 25℃; its optimal pH is 4.8-5.2, and its activity drops sharply below pH 4.0 or above 7.5. Protease has an optimal temperature of 40-45℃ and an optimal pH of 6.0-6.5. Considering the characteristics of all enzymes, the optimal temperature range is determined to be 25-55℃, and the optimal pH range is 4.5-7.0.

[0089] The effect of ionic strength is mainly reflected in its influence on enzyme conformation and substrate binding. Too low an ionic strength may cause enzyme molecules to unfold, reducing activity; too high an ionic strength may affect the formation of enzyme-substrate complexes, similarly reducing activity. Therefore, the ionic strength range was set at 0.05–0.2 mol / L.

[0090] To find the optimal combination of fermentation parameters, this method employs a hybrid particle swarm optimization-simulated annealing algorithm. This algorithm combines the global search capability of particle swarm optimization with the local fine-grained search capability of simulated annealing, making it particularly suitable for solving multi-objective nonlinear optimization problems. The specific implementation steps include:

[0091] Multiple particles are randomly generated within the fermentation parameter limits. Each particle contains three parameters: temperature, pH, and ionic strength. The fermentation parameter limits refer to the range between the maximum and minimum values ​​of the fermentation parameters. The default values ​​for the minimum and maximum temperatures are 25 and 55℃, respectively. The default values ​​for the minimum and maximum pH values ​​are 4.5 and 7.0, respectively. The default values ​​for the minimum and maximum ionic strength values ​​are 0.05 mol / L and 0.2 mol / L, respectively.

[0092] Particle initialization employs a Latin hypercube sampling method to ensure uniform distribution of initial particles within the parameter space. Each particle is represented by a three-dimensional vector: [temperature, pH, ionic strength]. For example, a particle might be [42.5℃, 5.8, 0.12 mol / L]. Initial velocities are randomly generated, with the velocity range set to 10% of the variation range of fermentation parameters; for example, the temperature velocity range is ±3℃, the pH velocity range is ±0.25, and the ionic strength velocity range is ±0.015 mol / L.

[0093] Calculate the overall fermentation efficiency value for each particle. The calculation process requires combining the temperature, pH value, and ionic strength of each particle to obtain environmental parameters. These environmental parameters, the current actual enzyme activity value, and the substrate concentration are then input into a trained environmental impact model. The actual enzyme activity value is the activity of each enzyme obtained from actual measurements, and the substrate concentration is obtained through the distribution of metabolites. The environmental impact model outputs the enzyme activity values ​​under these environmental parameters. These enzyme activity values ​​are then input into an enzyme interaction model to obtain the enzyme activity over a future period. Finally, the enzyme activity values ​​are input into a fermentation efficiency function, which outputs the fermentation efficiency.

[0094] Particles update their velocity and position based on their own optimal position and the global optimal position. For the i-th particle, the new position equals the current position plus the new velocity; the new velocity equals the inertia coefficient multiplied by the current velocity, plus the cognitive coefficient multiplied by the random number multiplied by the difference between the individual's optimal position and the current position, plus the social coefficient multiplied by the random number multiplied by the difference between the global optimal position and the current position.

[0095] The initial value of the inertia coefficient is set to 0.9, and it decreases linearly to 0.4 with the number of iterations. Specifically, the inertia coefficient is calculated as: 0.9 minus 0.5 multiplied by the current iteration number divided by the maximum iteration number. This design gives the algorithm strong global search capabilities in the early stages and strong local search capabilities in the later stages.

[0096] The cognitive coefficient and social coefficient are both set to 2.0, which is a classic parameter setting for particle swarm optimization, balancing the influence of individual and group experience. Random numbers are uniformly distributed in the interval [0,1] and are regenerated with each update, increasing the randomness and diversity of the algorithm.

[0097] The candidate solutions found by particle swarm optimization are locally searched using simulated annealing. The acceptance probability is dynamically adjusted based on the annealing temperature and the efficiency difference, which is the difference between the target efficiency and the actual efficiency. The optimal solution is the combination of fermentation parameters that maximizes the overall fermentation efficiency. Candidate solutions are combinations of fermentation parameters that may maximize the overall fermentation efficiency.

[0098] The annealing temperature update strategy employs exponential cooling. The next annealing temperature is equal to the cooling coefficient multiplied by the current temperature, where the cooling coefficient is set to 0.95. The initial temperature is set to 100, and the final temperature is set to 0.01. When the efficiency difference is negative, the annealing is accepted directly with a probability of 1. When the efficiency difference is positive, the acceptance probability is the negative efficiency difference divided by the power of the current annealing temperature (the natural constant).

[0099] The neighborhood search strategy employs Gaussian perturbation. Based on the current solution, Gaussian noise with a mean of 0 and a standard deviation equal to a preset multiple of the parameter range is added to each fermentation parameter to generate new fermentation parameters. Here, the current solution refers to the fermentation parameters among the particles currently being evaluated, the preset multiple has a default value of 5%, and the parameter range is the difference between the maximum and minimum values ​​of the fermentation parameter. For example, the standard deviation of the temperature parameter is the maximum temperature minus the minimum temperature multiplied by the preset multiple, which is 1.5℃.

[0100] To ensure all fermentation parameters remain within physical constraints, when a particle's position exceeds the upper boundary, it is set to the upper boundary value, and its velocity is reversed; conversely, when a particle's position exceeds the lower boundary, it is set to the lower boundary value, and its velocity is reversed. The upper boundary refers to the maximum value of the fermentation parameter, and the lower boundary refers to the minimum value. This approach guarantees that particles remain within the feasible region while maintaining the continuity of the search.

[0101] The process stops when the maximum number of iterations is reached, or when the efficiency difference falls below a set iteration threshold within a preset number of iterations, and outputs the globally optimal fermentation parameter combination. The globally optimal fermentation parameter combination refers to the fermentation parameters that maximize overall fermentation efficiency within the constraints of the fermentation parameters, including optimal temperature, optimal pH, and optimal ionic strength. The default maximum number of iterations is 150, and the default preset number is 20. The iteration threshold is defined as the improvement of the globally optimal solution being less than 0.1% within 20 consecutive iterations. The improvement is calculated as: the absolute value of the difference between the current optimal value and the previous optimal value, divided by the previous optimal value, and then multiplied by 100%.

[0102] In practical implementation, the initial number of particles was set to 20-30, the maximum number of iterations was 100-150, the inertia coefficient decreased linearly from 0.9 to 0.4, the cognitive coefficient and social coefficient were both set to 2.0, the initial simulated annealing temperature was set to 100°C, and the cooling coefficient was 0.95. These parameter settings were determined through multiple simulation tests, balancing the algorithm's convergence speed and search quality.

[0103] After optimization, the system will obtain a set of optimal fermentation parameter combinations (temperature, pH, and ionic strength) that maximize overall fermentation efficiency. This optimization result is directly transmitted to the real-time control system in the next step as parameter setpoints, and dynamically adjusted during fermentation based on real-time monitoring data to ensure the fermentation process remains in optimal condition.

[0104] The inputs to the environmental impact model are environmental parameters, the current actual enzyme activity value, and the substrate concentration. The output is the enzyme activity under the given environmental parameters, which include temperature, pH value, and ionic strength. The environmental impact model uses a deep neural network architecture to describe the degree of influence of changes in environmental parameters on the activities of various enzymes.

[0105] The environmental impact model employs a multi-layer feedforward neural network structure, comprising an input layer, multiple hidden layers, and an output layer. The input layer receives environmental parameters, current actual enzyme activity values, and substrate concentrations, which are then input into the network after feature standardization. The hidden layers consist of four fully connected layers, each containing 128 neurons. The ReLU activation function is used to enhance the model's non-linear expressive power. The number of neurons in the output layer equals the number of enzyme types, with each output node corresponding to a predicted activity value for one enzyme. The Sigmoid activation function ensures that the output value is within the range of 0-1.

[0106] The environmental impact model requires training data during training. This data collection includes: measuring enzyme activity every 2 degrees Celsius within a temperature range of 25-60 degrees Celsius under standard conditions of pH 6.5 and ionic strength of 0.1 mol / L, while simultaneously recording the current actual enzyme activity value and substrate concentration to obtain temperature-enzyme activity data pairs; measuring enzyme activity every 0.2 pH units within a pH range of 4.0-8.0 under standard conditions of 37 degrees Celsius and ionic strength of 0.1 mol / L, while simultaneously recording the current actual enzyme activity value and substrate concentration to obtain pH-enzyme activity data pairs; and measuring enzyme activity every 2 degrees Celsius within a temperature range of 25-60 degrees Celsius, while simultaneously recording the current actual enzyme activity value and substrate concentration to obtain pH-enzyme activity data pairs. Under standard conditions of 6.5, the activities of each enzyme were measured every 0.02 mol / L within the ionic strength range of 0.01-0.5 mol / L. The current true enzyme activity value and substrate concentration were recorded simultaneously to obtain ionic strength-enzyme activity data pairs. An orthogonal experiment was designed to simultaneously change three factors: temperature, pH, and ionic strength. The corresponding current true enzyme activity value and substrate concentration were recorded to obtain the mapping relationship data between environmental parameters, current true enzyme activity value, substrate concentration combination, and enzyme activity. This data was used to train the model to capture the interaction between parameters.

[0107] Environmental parameters, current actual enzyme activity values, and substrate concentrations were standardized, and enzyme activity data were normalized to the 0-1 range. Data augmentation techniques were employed, using interpolation methods to expand the training samples.

[0108] A mean squared error loss function is used, along with an L2 regularization term to prevent overfitting. The loss function is the sum of the squared errors between the predicted and actual values, plus the squares of the weight parameters, multiplied by the regularization coefficient.

[0109] The Adam optimizer was used with a learning rate of 0.001 and a batch size of 32. A learning rate decay strategy was employed during training, multiplying the learning rate by 0.9 every 50 training epochs.

[0110] Monitor the validation set loss, and stop training when the validation set loss does not decrease for 10 consecutive training epochs to prevent overfitting.

[0111] A 5-fold cross-validation method was used to evaluate model performance. The experimental data was randomly divided into 5 parts, with 4 parts used alternately as the training set and 1 part as the validation set. Evaluation metrics included root mean square error (RMSE), mean absolute error (MAO), and coefficient of determination (COP). The requirements were: RMSE less than 5%, MAO less than 3%, and COP greater than 0.92.

[0112] The SHAP (SHapley Additive exPlanations) method was used to analyze the contribution of various environmental parameters, current actual enzyme activity values, and substrate concentrations to the activities of different enzymes, verifying the biological rationality of the environment-enzyme activity relationship learned by the model.

[0113] Step 104: After obtaining the optimal combination of fermentation parameters, in order to cope with the dynamic changes in the fermentation process, calculate the parameter deviation between the current actual parameter value and the optimal combination of fermentation parameters, and calculate the amount of fermentation parameters that need to be adjusted based on the parameter deviation.

[0114] The specific process includes:

[0115] The actual fermentation parameters for the current fermentation process are obtained, including actual temperature, actual pH value, and actual ionic strength. The actual temperature is measured by temperature sensors distributed throughout the fermenter, with a measurement accuracy of ±0.1℃; the actual pH value is measured by a pH electrode, with a measurement accuracy of ±0.05℃; and the actual ionic strength is indirectly calculated by a conductivity sensor, with a measurement accuracy of ±0.01 mol / L.

[0116] Calculate the parameter deviations, which are the differences between the optimal and actual fermentation parameter values. The temperature deviation equals the optimal temperature minus the actual temperature; the pH deviation equals the optimal pH value minus the actual pH value; and the ionic strength deviation equals the optimal ionic strength minus the actual ionic strength. These deviations reflect the direction and magnitude of adjustments needed to the system.

[0117] The fermentation parameter adjustment strategy is determined based on the system's dynamic characteristics. Due to the inertia and delay characteristics of the fermentation system, the deviation cannot be directly used as the adjustment variable; the system's response time and stability requirements must be considered. The adjustment strategy employs a proportional-integral-derivative (PID) control algorithm, where the proportional term reflects the current deviation, the integral term eliminates steady-state error, and the derivative term predicts the deviation's trend. The fermentation parameters to be adjusted include temperature, pH, and ionic strength, where:

[0118] The temperature adjustment is equal to the proportional coefficient multiplied by the temperature deviation, plus the integral coefficient multiplied by the time integral of the temperature deviation, plus the derivative coefficient multiplied by the time derivative of the temperature deviation. The proportional coefficient is set to 0.6-0.8, the integral coefficient to 0.1-0.2 per minute, and the derivative coefficient to 2-3 minutes. These parameters are determined based on the thermal inertia characteristics and response speed requirements of the fermentation system. First, a system identification experiment is conducted, applying a step temperature input signal to the fermentation system and recording the system's temperature response curve to analyze the system's time constant and overshoot characteristics. Based on the response curve characteristics, the Ziegler-Nichols parameter tuning method is used to initially determine the PID parameters. Then, through on-site debugging and optimization, the proportional coefficient is gradually adjusted to obtain a suitable response speed, the integral coefficient is adjusted to eliminate steady-state error, and the derivative coefficient is adjusted to reduce overshoot and oscillation. The final parameter selection needs to balance the requirements of control accuracy, response speed, and system stability.

[0119] The calculation of pH adjustment needs to consider the influence of the buffer system. The pH adjustment equals the pH proportionality coefficient multiplied by the pH deviation, plus the pH integral coefficient multiplied by the time integral of the pH deviation, plus the pH derivative coefficient multiplied by the time derivative of the pH deviation. Due to the nonlinear characteristics of the pH system, the pH proportionality coefficient is set to 0.4-0.6, the pH integral coefficient to 0.05-0.1 per minute, and the pH derivative coefficient to 1-2 minutes. Determining the pH control parameters requires considering the buffering capacity of the fermentation broth and the response characteristics of the pH sensor. The buffering capacity of the fermentation broth is determined through titration experiments; a stronger buffering capacity requires a larger pH proportionality coefficient to overcome the buffering resistance. The setting of the pH integral coefficient needs to consider the delay characteristics of the pH system. Since the mixing of the pH adjuster takes time, the integral effect should not be too strong to avoid integral saturation. The selection of the pH derivative coefficient requires balancing anti-interference capability and noise amplification. By testing the control effect under different parameter combinations on-site, a parameter combination that achieves a pH control accuracy of ±0.05 without significant oscillations is selected.

[0120] The calculation of ion strength adjustment is relatively complex because ion strength adjustment is usually achieved by adding salt solution. The ion strength adjustment equals the ion strength proportional coefficient multiplied by the ion strength deviation, plus the ion strength integral coefficient multiplied by the time integral of the ion strength deviation, plus the ion strength derivative coefficient multiplied by the time derivative of the ion strength deviation. The ion strength proportional coefficient is set to 0.3-0.5, the ion strength integral coefficient to 0.02-0.05 per minute, and the ion strength derivative coefficient to 0.5-1 minute. The determination of ion strength control parameters is mainly based on the response characteristics of the salt solution addition system and the ion diffusion rate. Since ion strength adjustment is achieved by adding salt solution, its response speed is relatively slow; therefore, the ion strength proportional coefficient is set relatively small to avoid over-adjustment. The ion strength integral coefficient needs to consider the time constant of ion diffusion, typically 20-30 minutes; therefore, the integral action needs to be strong enough to eliminate steady-state deviations, but not too strong to avoid oscillations. The setting of the differential coefficient of ion intensity needs to take into account the noise level of ion intensity measurement and the inertia of the system. The optimal parameter combination is determined through step response experiments and frequency domain analysis to ensure that the ion intensity control accuracy reaches the requirement of ±0.005 mol / L.

[0121] The calculated adjustment amounts are limited and smoothed. Limiting the adjustment amounts prevents excessive adjustments from impacting the system, while smoothing avoids frequent parameter changes from affecting fermentation stability.

[0122] The temperature adjustment is limited to ±2℃ per adjustment, and the adjustment rate is limited to 0.5℃ / min. When the calculated temperature adjustment exceeds the limit, the limit is used as the actual adjustment, and the remaining deviation is processed in the next control cycle.

[0123] pH adjustments should be limited to ±0.3 per adjustment, with an adjustment rate not exceeding 0.1 / min. Special care must be taken when adjusting pH, as rapid pH changes can cause irreversible damage to microorganisms and enzyme activity.

[0124] The adjustment of ionic strength is limited to ±0.02 mol / L per adjustment, and the adjustment rate should not exceed 0.005 (mol / L) / min. Ionic strength adjustment is typically achieved by adding a pre-prepared salt solution, and the mixing time of the solution must be taken into account.

[0125] The smoothing process uses a first-order filter. The adjusted value after filtering is equal to the filter coefficient multiplied by the currently calculated adjusted value, plus 1 and minus the filter coefficient multiplied by the previous adjusted value. The filter coefficient is set to 0.7-0.9, which ensures both response speed and avoids excessive oscillation.

[0126] The processed adjustment is converted into actuator control signals. Temperature adjustment is achieved through heaters and coolers. The heater power adjustment equals the temperature adjustment multiplied by the heater power coefficient, and the cooler power adjustment equals the negative temperature adjustment multiplied by the cooler power coefficient. The heater and cooler power coefficients are determined through heat balance calculations, with typical values ​​of 50-100 W / ℃.

[0127] pH adjustment is achieved by adding acidic or alkaline solutions. The amount of acid solution added is equal to the amount of negative pH adjustment multiplied by the acid solution addition coefficient, and the amount of alkaline solution added is equal to the amount of positive pH adjustment multiplied by the alkaline solution addition coefficient. The addition coefficient is determined through titration experiments, taking into account the buffering capacity of the fermentation broth, and a typical value is 10-20 mL / pH unit.

[0128] Ionic strength adjustment is achieved by adding salt solution. The amount of salt solution added is equal to the amount of ionic strength adjustment multiplied by the fermentation broth volume and then divided by the salt solution concentration. For example, for 1000L of fermentation broth, to increase the ionic strength by 0.01 mol / L, using a 1 mol / L NaCl solution, 10L of solution needs to be added.

[0129] The actuator control signal output uses a standard 4-20mA current signal or 0-10V voltage signal, which is converted from digital control quantity to analog signal through a digital-to-analog converter. The control signal is updated every 30 seconds, which ensures timely control while avoiding overly frequent adjustments.

[0130] To ensure the safety and reliability of control, the system is equipped with multiple protection mechanisms. When a sensor malfunctions or the measured value is abnormal, the system automatically switches to safety mode, stops adjusting fermentation parameters, and issues an alarm. When an actuator malfunctions, the system automatically switches to a backup actuator or manual control mode. When parameter deviations are too large or the adjustment effect is unsatisfactory, the system will recalculate the optimal fermentation parameters or initiate a fault diagnosis program.

[0131] Through the precise calculation and control execution process of the aforementioned fermentation parameter adjustments, the system can accurately control the fermentation parameters within the optimal range, achieving automated and optimized control of the fermentation process, and improving fermentation efficiency and product quality stability. Practical applications show that this control method can improve temperature control accuracy to ±0.2℃, pH control accuracy to ±0.05, and ionic strength control accuracy to ±0.005 mol / L, significantly improving the control effect of the fermentation process.

[0132] Specific implementation methods include:

[0133] First, the target fermentation efficiency needs to be determined. This target efficiency is based on the fermentation efficiency of historically high-performing batches and is set at 95-98% of that historically high-performing batch efficiency. In actual production, this value can be obtained by analyzing the best-performing 10-15 batches from the historically high-performing batches, calculating their average fermentation efficiency, and then multiplying it by a coefficient of 0.95-0.98, ensuring that the target is both challenging and feasible.

[0134] The choice of a 95-98% coefficient is based on risk management considerations: setting targets that are too high will cause the system to excessively pursue extreme values ​​and neglect stability; setting targets that are too low will fail to fully realize the system's potential. The 95-98% range ensures that the target is challenging while leaving an appropriate safety margin to accommodate uncertainties such as raw material fluctuations and equipment aging.

[0135] The compensation model is the core component of real-time control. It compares the difference between the target efficiency and the measured efficiency to calculate the amount of parameters that need to be adjusted in the optimal fermentation parameters. Specifically, when the compensation model detects a deviation between the measured fermentation efficiency and the target efficiency, the compensation function maps the efficiency deviation to adjustments in fermentation parameters such as temperature, pH, and ionic strength. This mapping relationship is obtained through system identification methods, namely, analyzing the correspondence between changes in fermentation parameters and efficiency responses in historical data to establish a nonlinear regression model.

[0136] The specific implementation of system identification includes: collecting records of fermentation parameter adjustments and corresponding efficiency change data during historical fermentation processes; establishing a parameter-efficiency response model using the least squares method; evaluating the model accuracy through cross-validation; and periodically updating model parameters to adapt to system changes.

[0137] The efficiency deviation is calculated as follows: the deviation equals the difference between the target efficiency and the measured efficiency, divided by the target efficiency, and then multiplied by 100%. The measured efficiency is calculated in real time using the fermentation efficiency function defined in step 103, with a calculation frequency of once every 10 minutes to ensure timely detection of efficiency changes.

[0138] In practical applications, the compensation function adopts a piecewise linear or quadratic function form. When the efficiency deviation is small (<5%), linear compensation is used; when the deviation is large (≥5%), nonlinear enhanced compensation is used to ensure that the system can accurately respond to small deviations and quickly correct large deviations. For example, when the measured efficiency is 3% lower than the target efficiency, the temperature parameter may need to be increased by 1-1.5℃, and the pH value adjusted by 0.1-0.2 units.

[0139] The piecewise compensation function is designed based on nonlinear control strategies in control theory. The compensation coefficient for the linear compensation interval (deviation < 5%) is determined through linear regression; for example, the temperature compensation coefficient is 0.5℃ / %, meaning that for every 1% increase in efficiency deviation, the temperature is increased by 0.5℃. The nonlinear compensation interval (deviation ≥ 5%) adopts a quadratic function form, where the compensation intensity increases rapidly with increasing deviation, avoiding insufficient response under large deviations.

[0140] Specific methods for determining compensation coefficients include: analyzing the optimal adjustment amount under different deviation amplitudes in historical data; verifying the compensation effect through simulation experiments; conducting small-scale tests and fine-tuning in actual production; establishing a compensation effect evaluation mechanism and regularly optimizing compensation parameters.

[0141] The fermentation parameters were adjusted using a PID controller, with the proportional coefficient set to 0.8-1.2, the integral coefficient to 0.1-0.3, and the derivative coefficient to 0.05-0.15. These parameters were initially determined using the Ziegler-Nichols method and then fine-tuned based on the actual control effect.

[0142] The tuning process of PID controller parameters includes: first, determining the critical proportional coefficient and critical oscillation period using the critical proportionality method; then, calculating the initial PID controller parameters according to the Ziegler-Nichols formula; next, conducting a step response test in an actual system to observe the dynamic characteristics of the system; and finally, fine-tuning the parameters based on indicators such as overshoot, settling time, and steady-state error.

[0143] The selection of the integral time window for a PID controller requires a balance between the accuracy of trend identification and the timeliness of response. A window that is too short (<30 minutes) may be susceptible to random fluctuations, leading to incorrect trend judgments; a window that is too long (>60 minutes) may delay the identification of the true trend. By analyzing the characteristic cycles of efficiency changes in historical data, a window range of 30-60 minutes was determined to be optimal.

[0144] Determining the control parameters for these controllers requires a combination of system simulation and actual debugging: first, a mathematical model of the fermentation process is established, and the control effects of different combinations of fermentation parameters are tested in a simulation environment; then, verification and fine-tuning are performed in small-scale experiments; finally, final confirmation is made in actual production. Five to eight iterations of optimization are needed to determine the optimal combination of fermentation parameters.

[0145] The simulation model was established based on an enzyme interaction model, combined with a mass and heat transfer model and a microbial growth model, to construct a complete fermentation process simulation platform. Simulation verification included: steady-state response testing, dynamic response testing, anti-interference capability testing, and parameter sensitivity analysis.

[0146] Small-scale experiments were conducted in a 100L fermenter, with the experimental period being one complete fermentation process (48-72 hours). The experimental content included: system response characteristics under different control parameters, control accuracy and stability evaluation, and system performance under abnormal operating conditions.

[0147] The implementation of a real-time control system requires high-quality sensor data. The system employs a redundant measurement strategy, configuring 2-3 independent sensors for each key parameter. By fusing multi-sensor data through median filtering or weighted averaging algorithms, measurement reliability is significantly improved. Simultaneously, the system also implements automated sensor anomaly detection. When anomalies are detected, it automatically switches to backup sensors or uses historical data interpolation to ensure the continuous and stable operation of the control system.

[0148] The specific scheme for sensor redundancy configuration includes: temperature measurement using three PT100 platinum resistance temperature sensors, distributed in different locations in the fermenter; pH measurement using two glass electrode pH sensors, which are periodically calibrated with standard buffer solution; and ionic strength measurement using two conductivity sensors, which are calculated through the conductivity-ionic strength conversion relationship.

[0149] The selection of data fusion algorithms is based on sensor characteristics and measurement environment: for sensors with similar accuracy, simple averaging or weighted averaging is used; for sensors with large differences in accuracy, Kalman filtering or Bayesian fusion methods are used; for sensors with systematic bias, bias correction is performed before fusion.

[0150] Sensor anomaly detection employs multiple methods: range check (whether the value is within a reasonable range), rate of change check (whether the rate of change is abnormal), consistency check (whether the readings of multiple sensors are consistent), and trend check (whether the trend of change conforms to physical laws). When an anomaly is detected, the system will record the anomaly information, switch to a backup sensor, issue an alarm signal, and initiate a data interpolation program.

[0151] Data interpolation methods include: linear interpolation (suitable for short-term data loss), polynomial interpolation (suitable for smoothly changing parameters), model-based predictive interpolation (suitable for longer-term data loss), and parametric interpolation based on relevant parameters (using other parameters to infer missing parameters).

[0152] Enzyme activity monitoring and replenishment is a crucial component of this step. The system monitors the activity levels of various enzymes in real time, and automatically activates an enzyme replenishment decision algorithm when enzyme activity falls below a preset threshold. Enzyme activity monitoring employs a combination of online biosensors and indirect indicators, with a monitoring frequency of once every 30 minutes.

[0153] The enzyme activity thresholds are set based on the enzyme interaction model established in step 102, and are determined by analyzing the relationship between the activities of each enzyme and the overall fermentation efficiency. Specific thresholds include: triggering the corresponding enzyme replenishment decision algorithm when cellulase activity falls below 70% of its initial value, protease activity falls below 75% of its initial value, amylase activity falls below 80% of its initial value, and phytase activity falls below 65% of its initial value.

[0154] The enzyme replenishment decision algorithm comprehensively considers factors such as the current enzyme activity level, fermentation stage, remaining raw material quantity, and cost-effectiveness. The decision-making process of the enzyme replenishment algorithm is as follows: First, determine if the enzyme activity is below a threshold; determine the current fermentation stage, where enzyme replenishment is suitable in the early and middle stages of fermentation, but not in the later stages; calculate the ratio of the economic benefit of increased fermentation efficiency to the cost of the enzyme preparation when replenishing the enzyme; if the ratio exceeds a preset value (default value is 1.5), enzyme replenishment is performed; finally, determine the type and quantity of enzyme to be replenished.

[0155] The fermentation stage is determined based on the fermentation time and the changing trends of key indicators: in the early stage of fermentation (0-24 hours), the substrate is mainly consumed, and it is suitable to supplement various enzymes; in the middle stage of fermentation (24-48 hours), the product is mainly accumulated, and the focus is on supplementing restriction enzymes; in the late stage of fermentation (after 48 hours), the product is mainly stable, and it is generally not recommended to supplement enzymes.

[0156] The calculation of enzyme replenishment amount is based on the enzyme activity decay model and the target activity level. The calculation formula is: replenishment amount equals (target activity minus current activity) multiplied by fermentation volume, then divided by the enzyme activity concentration. The target activity is set at 85-90% of the initial activity to ensure that the enzyme activity is maintained at a high level.

[0157] The enzyme preparation should be added in small batches to avoid disrupting the fermentation environment with a large single addition. Each addition should not exceed 30% of the total required amount, with an interval of 2-4 hours between additions. Adjust the subsequent additions based on the recovery of enzyme activity. Before addition, the enzyme preparation needs to be diluted to an appropriate concentration and adjusted to a temperature and pH value close to that of the fermentation broth.

[0158] The enzyme preparation should be diluted using sterile physiological saline or fermentation medium at a ratio of 1:5 to 1:10. The diluted enzyme preparation must be added within 30 minutes to avoid loss of enzyme activity. It should be added continuously using a peristaltic pump at a rate of 0.1-0.2% of the fermentation volume per minute to ensure thorough mixing.

[0159] The effectiveness of enzyme supplementation is evaluated by comparing changes in fermentation efficiency before and after supplementation. Evaluation indicators include: the degree of enzyme activity recovery (the ratio of enzyme activity after supplementation to the target activity), the magnitude of fermentation efficiency improvement (the difference between efficiency after supplementation and efficiency before supplementation), and the cost-benefit ratio (the ratio of the economic benefit from efficiency improvement to the cost of the enzyme preparation). The specific calculation method for the cost-benefit ratio is as follows: First, calculate the economic benefit brought about by the increase in fermentation efficiency after supplementation, which can be obtained by multiplying the magnitude of the improvement in fermentation efficiency by the economic value corresponding to each unit of efficiency. Then, compare this economic benefit with the cost of the enzyme preparation to obtain the cost-benefit ratio. When the cost-benefit ratio is greater than 1.5, enzyme supplementation is considered economically reasonable.

[0160] The enzyme replenishment recording system automatically records detailed information for each enzyme replenishment, including: replenishment time, enzyme type, replenishment amount, enzyme activity levels before and after replenishment, changes in fermentation efficiency, and cost. These records are used to optimize enzyme replenishment strategies, establish an enzyme replenishment knowledge base, and provide a reference for subsequent batches.

[0161] By combining a compensation model with an enzyme activity monitoring and supplementation mechanism, control parameters and enzyme content can be adjusted in real time during fermentation, ensuring that fermentation efficiency remains near the target level. This effectively addresses the impact of uncertainties such as raw material changes, environmental disturbances, and enzyme activity decay, thereby improving the stability of the fermentation process and the consistency of product quality. Practice shows that after adopting this method, the batch-to-batch coefficient of variation in fermentation efficiency decreased from 12-15% under traditional control to 3-5%, while enzyme utilization efficiency increased by 15-20%, significantly improving production efficiency and product quality stability.

[0162] The batch-to-batch coefficient of variation (COP) is calculated as follows: Fermentation efficiency data from 100 consecutive batches are collected, the mean and standard deviation are calculated, and the COP is equal to the standard deviation divided by the mean and then multiplied by 100%. The COP of the traditional control method is obtained by analyzing historical data before adopting this method, while the COP of the new method is calculated from production data after implementing this method. The improved enzyme utilization efficiency is mainly reflected in the reduction of enzyme waste and the increase in effective enzyme utilization through precise enzyme replenishment strategies.

[0163] Example 2

[0164] Based on the basic control framework established in Example 1, and considering the spatial non-uniformity problem in large-scale fermentation systems, this example provides a multi-center adaptive control method, such as... Figure 3 As shown. This method extends the single-point control strategy of Example 1 to multi-point coordinated control. Through spatial domain partitioning and inter-domain coordination mechanisms, it solves the spatial non-uniformity problem of large-scale fermentation systems, further improving control accuracy and system stability.

[0165] Spatial inhomogeneity refers to the phenomenon in large-scale fermentation equipment where, due to factors such as limited mass and heat transfer, insufficient stirring, and complex geometry, parameters such as temperature, pH, dissolved oxygen concentration, and nutrient concentration vary significantly at different locations within the fermenter. This inhomogeneity leads to inconsistent fermentation results and affects the stability of product quality.

[0166] Large-scale fermentation systems refer to industrial fermentation equipment with a volume greater than 20 cubic meters. In equipment of this scale, the fluid mixing time constant is 10-30 minutes, while the characteristic time of the fermentation reaction ranges from several minutes to several hours. Therefore, insufficient mixing becomes the main factor affecting fermentation uniformity. In addition, the length-to-diameter ratio of large-scale equipment is 2-4, resulting in less thorough axial mixing than radial mixing, further exacerbating spatial inhomogeneity.

[0167] The specific steps include:

[0168] Step 201 involves dividing the fermentation system into multiple spatial domains, each equipped with independent sensors and actuators to achieve local parameter monitoring and control. The spatial domain division is based on the fermenter's geometry and fluid dynamics characteristics, ensuring relatively uniform parameters within each domain. Through rational spatial division, the complex, large system is decomposed into multiple relatively simple subsystems, facilitating precise control.

[0169] The total number of spatial domains is determined based on the size of the fermentation equipment and the fluid mixing characteristics. Industrial-scale fermenters (>20 cubic meters) are divided into 5-8 spatial domains. Domain division is primarily based on computational fluid dynamics simulation results, using fluid velocity, concentration, and temperature field analysis to determine regions with relatively uniform mixing. To ensure control effectiveness, the coefficient of variation of parameters within each spatial domain is required to be less than 5%, thus ensuring both domain uniformity and the complexity of the control system.

[0170] The specific implementation of computational fluid dynamics simulation includes: establishing a three-dimensional geometric model of the fermenter, including the tank body, agitator, baffles, inlet and outlet, etc.; setting boundary conditions, including agitator speed, feed flow rate, heat transfer coefficient, etc.; selecting a suitable turbulence model, such as the k-ε model or the Reynolds stress model; performing mesh generation, with a mesh size of 500,000 to 2 million; and solving the Navier-Stokes equations to obtain the flow field distribution.

[0171] Fluid velocity field analysis focuses on the distribution of velocity magnitude and direction, identifying high-velocity, low-velocity, and dead zones. Concentration field analysis simulates the distribution of nutrients or products by adding tracers. Temperature field analysis considers the generation and transfer of fermentation heat, identifying regions with large temperature gradients.

[0172] The requirement of a parameter coefficient of variation of less than 5% is determined based on sensitivity analysis of parameter uniformity during fermentation. Studies have shown that when the coefficient of variation of temperature within a region exceeds 5%, the difference in enzyme activity at different locations can reach 20-30%, severely affecting fermentation efficiency. Therefore, a coefficient of variation of 5% is a basic requirement for ensuring fermentation quality.

[0173] In the specific implementation process, a three-dimensional model of the fermentation tank is first established. Then, computational fluid dynamics software such as ANSYS Fluent or COMSOL Multiphysics is used to perform fluid dynamics simulations and obtain flow field distribution data. Based on the simulation results, the tank is automatically divided into several regions using K-means clustering or watershed segmentation algorithms. Finally, the division results are manually adjusted according to the actual engineering conditions to determine the final spatial domain distribution.

[0174] Applications of the K-means clustering algorithm include: using each grid point obtained from computational fluid dynamics simulation as a data point, and using parameters such as velocity, temperature, and concentration of that point as feature vectors; setting the number of clusters to 5-8; and iteratively optimizing to group grid points with similar features into the same class; with each class corresponding to a spatial domain.

[0175] Applications of the watershed segmentation algorithm include: using the parameter gradient as "terrain height", with areas of large gradient corresponding to "ridges" and areas of small gradient corresponding to "valleys"; starting from the point of smallest gradient to "fill water", forming different "watersheds"; each "watershed" corresponds to a spatial domain.

[0176] Manual adjustments primarily consider practical engineering factors: installation location limitations for sensors and actuators; the impact of piping and structural components on the spatial domain; ease of maintenance and operation; and a balance between cost and complexity. The adjusted spatial domain should meet the following requirements: a relatively regular shape to facilitate sensor placement; a relatively balanced volume to avoid uneven control loads; and clear boundaries to reduce inter-domain interference.

[0177] Step 202 involves performing local optimal control for each spatial domain. The optimal combination of fermentation parameters (temperature, pH, and ionic strength) for that domain is calculated using the methods described in steps 101 to 104 of Example 1, and the enzyme content is supplemented. This step ensures that optimal fermentation results are achieved within each spatial domain.

[0178] For each spatial domain, the system collects data from all sensors within that region, calculating the average and variance of the parameters within that region. Based on this data, a mathematical model of the region is established, and the optimal parameter combination is solved using a particle swarm optimization-simulated annealing hybrid algorithm.

[0179] The sensor placement strategy includes: at least 3 temperature sensors, 2 pH sensors, 2 dissolved oxygen sensors, and 1 ion intensity sensor in each spatial domain; the sensor locations should avoid the stirrer sweep area and dead zone area; the distance between sensors should ensure the representativeness of the measurement and not exceed 1 / 3 of the maximum size in the domain.

[0180] The data processing methods include: weighted averaging of data from multiple sensors within the same domain, with the weights determined based on the representativeness of the sensor locations; identifying and removing outlier data using the Laida criterion or box plot method; and performing time filtering on the data to eliminate the impact of short-term fluctuations.

[0181] The establishment of a regional mathematical model needs to take into account the particularities of the region: the fluid mixing characteristics within the region are determined through computational fluid dynamics simulation or tracer experiments; the heat transfer characteristics within the region are considered in terms of heat exchange with adjacent regions and the outside world; and the biochemical reaction characteristics within the region may vary depending on local conditions.

[0182] In this process, the fermentation efficiency of each spatial domain is calculated by weighted averaging of the aforementioned indicators such as sugar consumption rate, product formation rate, and enzyme activity. These indicators are all obtained by real-time data acquisition from sensors within that region. Specifically, the sugar consumption rate is calculated by measuring the rate of decrease from time-series data from a glucose concentration sensor; the product formation rate is calculated by measuring the rate of increase from time-series data from a specific product concentration sensor; and enzyme activity is assessed using dedicated biosensors or indirect indicators (such as the consumption rate of a specific substrate).

[0183] The sugar consumption rate was determined using the glucose oxidase electrode method, which features rapid response, high accuracy, and strong anti-interference capability. The sensor collected data every 5 minutes, and the consumption rate was obtained by calculating the concentration difference between two consecutive time points divided by the time interval. During normal fermentation, the sugar consumption rate was 0.5–2.0 g / L / h.

[0184] The product formation rate was determined using appropriate detection methods depending on the target product: organic acids were analyzed using ion chromatography or near-infrared spectroscopy; amino acids were analyzed using an amino acid analyzer or fluorescence detection; and proteins were analyzed using the biuret method or the Bradford method. The detection frequency was once every 30 minutes, and the formation rate was obtained by fitting the slope of the concentration-time curve.

[0185] Indirect methods for measuring enzyme activity include: estimating enzyme activity from the rate of consumption of a specific substrate, such as estimating cellulase activity from the rate of cellulose degradation; estimating enzyme activity from the rate of product formation, such as estimating protease activity from the rate of amino acid formation; and estimating total enzyme activity from the rate of respiration, such as assessing metabolic activity from the rate of oxygen consumption or the rate of carbon dioxide production.

[0186] Step 203: To address the potential global incoordination issues caused by local optimization within a domain and to prevent system instability due to excessive differences in parameters between adjacent spatial domains, an inter-domain coordination function mechanism is established. By coordinating the parameter differences between each spatial domain, the coordinated and stable operation of the entire system is ensured.

[0187] This coordination function is implemented through a mathematical function that considers the parameter differences between adjacent domains and their physical distance. Specifically, the function calculates the differences in control parameters (temperature, pH, and ionic strength) between two spatial domains and weights them according to their physical distance. When the two domains are close and the parameter differences are large, the system exerts a stronger coordination effect; when the two domains are far apart or the parameter differences are small, the coordination effect is weaker.

[0188] The physical basis for inter-domain coordination is mass and heat transfer processes. There is an exchange of matter and energy between adjacent domains, and excessively large parameter differences can lead to strong driving forces for mass and heat transfer, potentially causing system instability. For example, excessively large temperature differences between adjacent domains can lead to strong convective heat transfer, affecting the stability of temperature control; excessively large pH differences can lead to ion diffusion, affecting local ion balance.

[0189] The mathematical expression for the coordination function is: the coordination effect equals the coordination intensity coefficient multiplied by the parameter difference, then multiplied by the negative spatial attenuation coefficient of the natural constant, multiplied by the distance raised to the power of the power. Here, the parameter difference is calculated using normalized Euclidean distance, which is the square root of the sum of the squares of the parameter differences divided by the parameter range; the distance is the straight-line distance between the geometric centers of the two domains; and the exponential function reflects the attenuation effect of distance on the coordination effect.

[0190] The coordination strength coefficient and the spatial decay coefficient are two key parameters, determined through system dynamic response analysis. Typical values ​​are a coordination strength coefficient of 0.4–0.6 and a spatial decay coefficient of 0.2–0.3 per meter. These parameters were obtained through small-scale experimental verification of the fermentation system and optimization via computer simulation. The system was configured with parameter difference thresholds: temperature not exceeding 2°C, pH not exceeding 0.3, and ionic strength not exceeding 0.02 mol / L. When the parameter differences between adjacent domains exceed these thresholds, a stronger coordination effect is triggered.

[0191] The methods for determining the coordination intensity coefficient include: analyzing the time constants of mass and heat transfer between domains to determine the intensity requirement of the coordination effect; testing the system stability under different coordination intensities through simulation experiments; and optimizing parameters in a real system to find the optimal coordination intensity. Too low a coordination intensity may not effectively suppress inter-domain differences, while too high a coordination intensity may weaken the effect of local optimization.

[0192] The spatial attenuation coefficient reflects the degree to which distance affects the coordination effect. This coefficient is determined based on the physical laws of mass and heat transfer; the driving force of mass and heat transfer is proportional to the concentration gradient or temperature gradient, and the influence of the gradient decreases with increasing distance. A reasonable range for the attenuation coefficient is determined by analyzing the mass and heat transfer characteristics within the fermenter.

[0193] The parameter difference thresholds were set based on the stability requirements of the fermentation process. Temperature differences exceeding 2°C may lead to significant differences in enzyme activity, affecting fermentation uniformity; pH ​​differences exceeding 0.3 may affect microbial growth and enzyme activity; and ionic strength differences exceeding 0.02 mol / L may affect protein conformation and enzyme activity. These thresholds were determined through experimental verification and literature review.

[0194] Physical distance is calculated as the straight-line distance between the geometric centers of the spatial domains. In practical engineering, it is obtained by setting up a three-dimensional coordinate system for the fermenter and calculating the Euclidean distance between the center points of each spatial domain.

[0195] For regularly shaped spatial domains, the geometric center is used directly; for irregularly shaped spatial domains, the centroid calculation method is used, which is a weighted average of the coordinates of each grid point, with the weight being the grid volume. The coordinate system is established with the center of the bottom of the fermenter as the origin, the axial direction as the z-axis, and the radial direction as the xy-plane.

[0196] Step 204: Combining the results of step 202 and the inter-domain coordination mechanism of step 203, a multi-domain joint optimization objective is established. This objective aims to maximize overall fermentation efficiency while minimizing inter-domain parameter inconsistencies. This step is the core algorithmic step of multi-center control, achieving an organic unity between local optimization and global coordination.

[0197] The system constructs a comprehensive optimization objective function, which consists of two parts: first, the sum of fermentation efficiencies across all spatial domains, representing the overall system performance; and second, the sum of coordination function values ​​between all possible pairs of spatial domains, representing the system's coordination. The system balances local optimization and global coordination by maximizing the difference between the sum of efficiencies and the coordination function.

[0198] The mathematical expression for the overall optimization objective function is: the objective function equals the sum of fermentation efficiencies across all domains minus the sum of the coordination weight coefficients multiplied by the coordination functions between all domains, where the coordination weight coefficients are used to balance the importance of efficiency and coordination. The coordination weight coefficients range from 0.1 to 0.3; too small a value may lead to insufficient coordination, while too large a value may result in excessive sacrifice of efficiency.

[0199] Methods for determining the coordination weight coefficients include: using weight determination methods from multi-objective optimization theory, such as the analytic hierarchy process (AHP) or expert scoring; testing system performance under different weights through simulation experiments; and verifying the rationality of the weights using actual production data. The final determined weights should maximize overall efficiency while ensuring a certain level of coordination within the system.

[0200] The calculation of the inter-domain coordination function value needs to consider all possible combinations of domain pairs. For n spatial domains, there are n multiplied by n minus 1 and then divided by 2 domain pairs. The coordination function value of each domain pair is calculated according to the coordination function defined in step 203. The sum of the coordination function values ​​of all domain pairs reflects the overall coordination degree of the system.

[0201] Solving this multi-objective optimization problem employs a distributed computing framework. First, each spatial domain controller computes local optima in parallel. Then, a central coordinator integrates these solutions and applies a coordination function to adjust the parameters. This process iterates until convergence. The convergence criterion is that the parameter changes are below preset thresholds for three consecutive iterations, including temperature less than 0.2℃, pH less than 0.05, and ionic strength less than 0.005 mol / L; or the maximum number of iterations is reached.

[0202] The specific implementation of the distributed computing framework includes: each spatial domain controller is equipped with an independent computing unit, capable of executing local optimization algorithms; the central coordinator is responsible for collecting the optimization results of each domain, calculating the coordination adjustment amount, and sending the adjusted parameters back to each domain controller; the communication network adopts industrial Ethernet or fieldbus to ensure the real-time performance and reliability of data transmission.

[0203] Simplified optimization algorithms, such as gradient descent or the simplex method, are used to calculate local optima to reduce computation time. The coordination algorithm of the central coordinator uses the Lagrange multiplier method or the penalty function method to transform the constrained optimization problem into an unconstrained optimization problem.

[0204] The specific steps of the iterative process include: each domain controller calculates the local optimal parameters based on the current state; the central coordinator collects the optimization results of all domains; the inter-domain coordination function value and the overall objective function value are calculated; the parameters of each domain are adjusted according to the coordination mechanism; the convergence condition is checked to decide whether to continue iterating; and the final parameters are sent to each domain executor.

[0205] The convergence threshold is set based on considerations of control accuracy requirements and computational efficiency. An overly strict threshold may lead to excessively long convergence times, affecting real-time control performance; an overly lenient threshold may result in insufficient control accuracy. A suitable threshold range is determined through simulation experiments and actual testing.

[0206] The maximum number of iterations is set based on computation time constraints. The total time for each coordination optimization should be controlled within 10-15 minutes to ensure real-time control requirements. A reasonable maximum number of iterations is determined based on the computing power and communication latency of each domain controller.

[0207] Through this coordinated optimization process, the system can obtain a control scheme that satisfies both the local optimal requirements of each region and maintains a smooth transition of overall parameters. This effectively solves the problem of spatial non-uniformity in large-scale fermentation systems, improving the stability of the fermentation process and the consistency of product quality. Practice has shown that after adopting this method, parameter deviations between different regions of a large-scale fermentation system have been reduced by 60-70%, and product quality consistency has been improved by 25-30%.

[0208] The calculation method for reducing parameter deviation is as follows: The standard deviation of parameter differences between domains before and after coordinated control is used to calculate the percentage reduction. The evaluation of product quality consistency uses the coefficient of variation of key product indicators (such as protein digestibility and nutrient content); a decrease in the coefficient of variation indicates improved consistency.

[0209] Example 3

[0210] See Figure 4 As shown, a feed fermentation parameter control system is provided, which stores computer-readable instructions and, when the computer-readable instructions are read, can execute the aforementioned feed fermentation parameter control method. The system includes:

[0211] The identification module 301 acquires the distribution of metabolites during the fermentation process and calculates the metabolome similarity index based on the distribution of metabolites.

[0212] Analysis module 302 establishes an enzyme activity correlation matrix, determines enzyme activity based on the distribution of metabolites, constructs an enzyme interaction model based on the enzyme activity correlation matrix and enzyme activity, and outputs the trend of changes in the activity of each enzyme over a future period of time.

[0213] The optimization module 303 combines the outputs of the metabolomics similarity index and the enzyme interaction model to define a fermentation efficiency function and solve the fermentation efficiency function to obtain the optimal parameter combination.

[0214] The control module 304 establishes a compensation model, compares the difference between the target efficiency and the measured efficiency, and calculates the amount of parameters that need to be adjusted in the optimal parameters; it also establishes an enzyme activity monitoring and replenishment mechanism, and when the enzyme activity is detected to be lower than the preset threshold, it initiates the enzyme replenishment decision algorithm.

[0215] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method of controlling a feed fermentation parameter, characterized by, The method comprises the following steps: Obtaining the metabolic product distribution in the feed fermentation process, and calculating the metabolome similarity index through the metabolic product distribution; including: obtaining the concentration of each metabolic product in the metabolic product distribution; extracting the reference concentration and weight coefficient of each metabolic product from the database; calculating the absolute difference between the concentration of each metabolic product and the reference concentration, and obtaining the relative deviation by dividing the absolute difference by the reference concentration; multiplying the relative deviation by the weight coefficient of the corresponding metabolic product to obtain the weighted deviation; summing the weighted deviations of multiple metabolic products to obtain the metabolome similarity index; Establishing an enzyme activity correlation matrix, determining enzyme activity according to the metabolic product distribution, and constructing an enzyme interaction model according to the enzyme activity correlation matrix and the enzyme activity, and the enzyme interaction model outputs the enzyme activity in the future period of time; Combining the metabolome similarity index and the enzyme activity in the future period of time, defining a fermentation efficiency function, and solving the fermentation efficiency function to obtain the optimal fermentation parameter combination; the fermentation efficiency function includes a weighted total enzyme activity term and a metabolome similarity contribution term; wherein the total enzyme activity term is the sum of the enzyme activity of each enzyme multiplied by the relative importance coefficient of each enzyme, and then multiplied by the preset total enzyme activity weight coefficient; the metabolome similarity contribution term is the metabolome similarity index multiplied by the negative value of the preset metabolome similarity weight coefficient; the metabolome similarity index and the enzyme activity are input into the fermentation efficiency function, and the fermentation efficiency function outputs the fermentation efficiency; the particle swarm simulated annealing hybrid algorithm is used to solve the fermentation efficiency function, and the steps include: Randomly generating a plurality of particles within the fermentation parameter limit range, each particle containing a fermentation parameter, wherein the fermentation parameter includes temperature, pH value and ionic strength, and the fermentation parameter limit range is the value range between the maximum and minimum values of the fermentation parameter; Calculating the fermentation efficiency value corresponding to each particle, including combining the temperature, pH value and ionic strength of each particle to obtain the environmental parameter, inputting the environmental parameter, the current real enzyme activity value and the substrate concentration into the trained environmental impact model, and the environmental impact model outputs the enzyme activity value, inputting the enzyme activity value into the enzyme interaction model to obtain the enzyme activity in the future period of time, and inputting the enzyme activity into the fermentation efficiency function to output the fermentation efficiency value; The particle updates the speed and position according to the best position of itself and the global best position; Setting an acceptance probability, applying the simulated annealing algorithm to locally search the particle swarm to find a candidate solution; the acceptance probability is adjusted according to the annealing temperature and the efficiency difference value; the efficiency difference value is the difference between the target efficiency and the actual efficiency; the candidate solution is the fermentation parameter combination that can maximize the overall fermentation efficiency value; the adjustment rule of the acceptance probability is that when the efficiency difference value is negative, the acceptance probability is 1, and when the efficiency difference value is positive, the acceptance probability is the negative efficiency difference value divided by the value of the current annealing temperature raised to the power of the natural constant; On the basis of the current solution, Gaussian noise with a mean value of 0 and a standard deviation of a preset multiple of the parameter range is added to each fermentation parameter to generate new fermentation parameters; wherein the current solution is the fermentation parameter in the particle being currently evaluated, and the parameter range is the difference between the maximum and minimum values of the fermentation parameter. When the maximum number of iterations is reached, or the efficiency difference is lower than the set iteration threshold in a preset number of consecutive iterations, the optimal solution is output, and the optimal solution is a combination of fermentation parameters that maximizes the overall fermentation efficiency value; Calculate the parameter deviation between the current actual parameter value and the optimal fermentation parameter combination, and calculate the amount of fermentation parameter adjustment according to the parameter deviation.

2. A feed fermentation parameter control method according to claim 1, characterized in that, The enzyme activity correlation matrix represents the interaction between all enzymes in a two-dimensional table, where each element represents the degree of influence of one enzyme on another enzyme, and the degree of influence is represented by an influence coefficient; the number of rows and columns of the enzyme activity correlation matrix is equal to the total number of enzymes, and each element in the enzyme activity correlation matrix represents the influence coefficient of the enzyme corresponding to the row on the enzyme corresponding to the column, a positive value represents a promoting effect, a negative value represents an inhibitory effect, and a zero value represents no interaction.

3. A feed fermentation parameter control method according to claim 2, characterized in that, The enzyme interaction model is a differential equation that describes the change of enzyme activity over time through a mathematical expression, which includes the growth of enzyme activity and the degradation of enzyme, where the growth of enzyme activity is the product of the inherent activity coefficient of the enzyme and the current enzyme activity, plus the sum of the product of the preset influence coefficient and each enzyme activity; the degradation of the enzyme is represented by the product of the degradation rate and the current enzyme activity.

4. The method of claim 1, wherein the step of determining the feed fermentation parameter comprises the steps of: determining a feed fermentation parameter based on the feed fermentation parameter model; and determining a feed fermentation parameter based on the feed fermentation parameter model. The relative importance coefficient of each enzyme is determined according to the influence effect of each enzyme on the metabolic product, and the influence effect is the average value of the percentage change in the distribution of the corresponding metabolic product caused by the increase of a certain enzyme activity in a preset range.

5. The method of claim 1, wherein the step of determining the feed fermentation parameter comprises the steps of: determining a feed fermentation parameter based on the feed fermentation parameter model; and determining a feed fermentation parameter based on the feed fermentation parameter model. The current actual parameter value is the actual fermentation parameter value of the current fermentation process, including the actual temperature, the actual pH value and the actual ionic strength; The parameter deviation is the difference between the optimal fermentation parameter combination and the actual fermentation parameter value, including the temperature deviation amount, the pH deviation amount and the ionic strength deviation amount, wherein the temperature deviation amount is equal to the optimal temperature minus the actual temperature; The pH deviation amount is equal to the optimal pH value minus the actual pH value; the ionic strength deviation amount is equal to the optimal ionic strength minus the actual ionic strength.

6. A feed fermentation parameter control method according to claim 5, wherein, The amount of fermentation parameter adjustment includes temperature adjustment amount, pH adjustment amount and ionic strength adjustment amount, wherein: The temperature adjustment amount is equal to the preset temperature proportion coefficient multiplied by the temperature deviation amount, plus the preset temperature integral coefficient multiplied by the time integral of the temperature deviation amount, plus the preset temperature differential coefficient multiplied by the time differential of the temperature deviation amount; The pH adjustment amount is equal to the preset pH proportion coefficient multiplied by the pH deviation amount, plus the preset pH integral coefficient multiplied by the time integral of the pH deviation amount, plus the preset pH differential coefficient multiplied by the time differential of the pH deviation amount; The ionic strength adjustment amount is equal to the preset ionic strength proportion coefficient multiplied by the ionic strength deviation amount, plus the preset ionic strength integral coefficient multiplied by the time integral of the ionic strength deviation amount, plus the preset ionic strength differential coefficient multiplied by the time differential of the ionic strength deviation amount.

7. A feed fermentation parameter control system, characterized by, It is used to perform a feed fermentation parameter control method as claimed in any one of claims 1-6, and the system comprises: The identification module acquires the metabolic product distribution in the feed fermentation process, and calculates the metabolome similarity index through the metabolic product distribution; The analysis module establishes an enzyme activity correlation matrix, determines enzyme activity according to metabolite distribution, constructs an enzyme interaction model according to the enzyme activity correlation matrix and enzyme activity, and the enzyme interaction model outputs each enzyme activity in a future period of time; The optimization module combines the metabolome similarity index and each enzyme activity in the future period of time, defines a fermentation efficiency function, and solves the fermentation efficiency function to obtain an optimal fermentation parameter combination; The control module calculates a parameter deviation between a current actual parameter value and the optimal fermentation parameter combination, and calculates an amount of fermentation parameters to be adjusted according to the parameter deviation.

Images